Types of Data Analysis: Descriptive, Diagnostic, Predictive, Prescriptive

Most teaching materials group data analysis into four types: descriptive (what happened), diagnostic (why it happened), predictive (what is likely next) and prescriptive (what to do). They are not four separate careers. They are four depths of question you can ask of the same dataset. This post explains each type, adds a short methods section for the "two methods" search queries, and works every example on our illustrative training-institute enquiry file.

I'm Gulab Chand Tejwani. I teach data analysis at TechNet Consultancy in Vallabh Vidyanagar, Anand, and online. In class I introduce the four types only after students can clean a file; otherwise the labels stay abstract.

The practice file behind the examples

All counts below come from the synthetic file technet-institute-enquiries-practice.csv (854 rows including deliberate duplicates; labelled illustrative). Downloads:

• CSV
• Data dictionary

After keeping the first row per enquiry_id, the file has 852 unique enquiry IDs (checked 4 November 2026). Status mix on the raw export: 300 Enquired, 212 Follow-up, 162 Enrolled, 135 Lost, 45 Duplicate.

1. Descriptive analysis — what happened?

Descriptive analysis summarises the past: counts, totals, averages, rates and simple breakdowns. No claim about causes yet.

Example (illustrative): after de-duplicating on enquiry_id, enrolments by source include:

Source	Enquiries	Enrolled	Conversion
Website form	107	26	24.3%
Walk-in	107	24	22.4%
Google	114	22	19.3%
Instagram	108	21	19.4%
(blank)	115	21	18.3%
Facebook	92	17	18.5%
College visit	109	17	15.6%
Referral	100	14	14.0%

Those figures answer "what happened in the pipeline by channel?" They do not yet say why Referral converts lower than Website form.

2. Diagnostic analysis — why did it happen?

Diagnostic analysis looks for explanations: comparisons, segments, funnels and quality checks.

Example (illustrative): Referral conversion is 14.0% and College visit is 15.6%, while Website form is 24.3% and Walk-in is 22.4% on the same de-duplicated file. One possible explanation is slower follow-up on referred leads; another is that college-visit enquiries are earlier in the decision cycle. The numbers alone do not pick the cause — they tell you which channels to investigate first (call logs, counsellor notes, time-to-first-contact).

Blank source is still a quality issue even when its conversion sits mid-pack: it is 115 / 852 unique enquiries (13.5%), so every channel ranking is partly incomplete until the form and import rules require a source.

Another diagnostic cut: among Follow-up rows with a numeric follow_up_days, the average is 10.7 days (n = 177, checked 4 November 2026). Per the data dictionary that field is days from enquiry to the last follow-up, not time to first call — so it cannot prove whether staff missed a three-day first-contact standard. It does show that many open leads are still being touched more than a week after the enquiry, which is worth investigating with a proper first-contact timestamp if you have one.

3. Predictive analysis — what is likely next?

Predictive analysis uses patterns in history to estimate future outcomes: who is likely to enrol, which leads may go cold, next month's enquiry volume.

Example (illustrative, deliberately simple): Website form and Walk-in convert higher than Referral in this file. A beginner predictive step is a rules-based score, not a neural network:

• +2 if source is Website form or Walk-in
• +1 if course is filled
• +1 if status is already Follow-up
• −2 if status is Lost or Duplicate
• −1 if follow_up_days is greater than 7 when present

Rank open leads by score and call the top of the list first. That is predictive in intent. A statistical model can come later once you have cleaner history and more months of data. Treat any score built on this synthetic file as practice only.

4. Prescriptive analysis — what should we do?

Prescriptive analysis recommends an action, sometimes with trade-offs.

Example (illustrative): given Referral at 14.0% conversion versus Website form at 24.3%, and an average 10.7 days from enquiry to last follow-up on Follow-up rows in this file, a prescriptive recommendation for a similar real CRM would be:

1. Make course and source required on the web form (13.5% of unique enquiries still lack a source here).
2. Record first-contact time separately from last follow-up, then build a daily list of leads that miss your first-call SLA.
3. Review the Referral hand-off script before spending more on that channel.

Prescription needs authority and cost awareness. "Spend twice the ads budget on Website form traffic" would be reckless from this sample alone; fixing data capture and measuring first contact properly are safer first prescriptions because they improve every later analysis.

Type	Question	Typical output
Descriptive	What happened?	Totals, rates, dashboards
Diagnostic	Why did it happen?	Segments, funnel leaks, quality checks
Predictive	What is likely next?	Scores, forecasts, risk lists
Prescriptive	What should we do?	Recommended actions and priorities

Methods of data analysis (qualitative and quantitative)

Search queries such as "two methods of data analysis" usually mean this split:

• Quantitative methods work with numbers you can count or measure: enrolment rates, fees, follow-up days, conversion.
• Qualitative methods work with meaning in text or conversation: interview notes, open feedback, the notes column in our practice file ("Wants weekend batch", "Asked for discount").

Both methods can support any of the four types. A descriptive summary of course demand can mix enrolment counts (quantitative) with themes from notes (qualitative). Diagnostic work often starts quantitative ("conversion fell") and finishes qualitative ("counsellors say weekend batches filled").

Collection techniques (surveys, interviews, observation, system extracts, web forms) feed the methods; they are not a fifth competing "type" list. See the collect step in data analysis process.

How to practise

1. Download the illustrative CSV and dictionary.
2. Reproduce the source conversion table in Excel or pandas.
3. Write one sentence each for descriptive, diagnostic, predictive and prescriptive views of the same table.
4. Only then change tools — SQL or Power BI — keeping the questions fixed.

For the end-to-end workflow and a free checklist, use data analysis process. For definitions and a smaller example, see what is data analysis.

If you want feedback on your first project, our data analysis training runs in Vallabh Vidyanagar, Anand, and online.

FAQ

What are the 4 types of data analysis?

Descriptive, diagnostic, predictive and prescriptive: what happened, why, what is likely next, and what to do.

What are the 3 types of data analysis?

Some primers stop at descriptive, diagnostic and predictive, and treat prescription as part of decision-making rather than analysis. The four-type list is the one I teach because action is part of the job.

What are two methods of data analysis?

Quantitative (numbers) and qualitative (meaning in text or conversation). They complement each other.

Which type should a beginner learn first?

Descriptive, then diagnostic, on a messy but labelled dataset. Predictive and prescriptive come after you trust your cleaning and definitions.